livebench.ai · benchmark source
LiveBench
A contamination-aware suite that refreshes questions over time. Category and subtask views live on the source site.
Fresh evaluation
LiveBench overall
A benchmark family designed to refresh questions and reduce contamination from training data.
Why it matters
Fresh tasks can provide a less stale view of reasoning and language performance.
Use it when
You want a current external signal alongside static benchmark suites.
The limitation
It still measures selected tasks and cannot predict your full workflow or tool integration.
Leaderboard snapshot
LiveBench overall results in a readable view
Use the tabs for the primary score and other published fields from this board. The source link remains authoritative for the live table.
Local leader
Claude Fable 5 Max Effort
83%
Rows shown
43
Overall score
Snapshot date
2026-06-25
51 days old · not a live API feed
Open the live source before trusting rank order — especially when this snapshot predates recent model releases.
Score profile
Overall score by published row
Higher is better in this view
Claude Fable 5 Max Effort
GPT-5.6 Sol Max EffortGPT-5.5 Thinking xHigh EffortClaude 5 Opus Thinking Max Effort
Smaug-Agentic
Kimi K3
Gemini 3.7 Flash High
Qwen 3.8 MaxGPT-5.4 Thinking xHigh Effort
Grok 4.6
Muse Spark 1.2 xHigh EffortGPT-5.6 Terra Max Effort
DeepSeek V4 Pro 0813Gemini 3.1 Pro Preview HighClaude 4.7 Opus Thinking xHigh EffortClaude 4.8 Opus Thinking Max EffortClaude Sonnet 5 xHigh EffortGrok 4.5Muse Spark 1.1 xHigh EffortGemini 3.5 Flash HighGPT-5.2 HighClaude 4.6 Opus Thinking High EffortDeepSeek V4 Flash 0731GPT-5.2 CodexGemini 3.6 Flash HighGPT-5.6 Luna Max Effort
GLM-5.2Qwen 3.7 MaxClaude 4.6 Sonnet Thinking Medium EffortClaude 4.5 Opus Thinking High Effort
Inkling xHigh EffortDeepSeek V4 ProKimi K2.6 ThinkingGPT-5.4 Nano xHighQwen 3.6 PlusKimi K2.7 CodeGrok Build 0.1
Minimax M3GPT-5.4 Mini xHighDeepSeek V4 FlashQwen 3.6 27BGemini 3.5 Flash-Lite HighGrok 4.3Source detail
Cost / task from the published rows
Sorted by this field · not the primary ranking
Claude Fable 5 Max EffortClaude 4.8 Opus Thinking Max EffortClaude 5 Opus Thinking Max EffortClaude 4.5 Opus Thinking High EffortClaude 4.7 Opus Thinking xHigh EffortGPT-5.6 Sol Max EffortClaude Sonnet 5 xHigh EffortGPT-5.5 Thinking xHigh EffortClaude 4.6 Opus Thinking High EffortGPT-5.4 Thinking xHigh EffortMuse Spark 1.2 xHigh EffortGPT-5.6 Terra Max EffortKimi K3GPT-5.4 Mini xHighSmaug-AgenticInkling xHigh EffortClaude 4.6 Sonnet Thinking Medium EffortGemini 3.1 Pro Preview HighQwen 3.8 MaxGemini 3.5 Flash HighGemini 3.6 Flash HighGPT-5.2 HighQwen 3.6 PlusGLM-5.2Grok 4.6Qwen 3.6 27BMuse Spark 1.1 xHigh EffortGPT-5.2 CodexQwen 3.7 MaxGPT-5.6 Luna Max EffortKimi K2.6 ThinkingGemini 3.7 Flash HighGrok 4.5Kimi K2.7 CodeGPT-5.4 Nano xHighGemini 3.5 Flash-Lite HighGrok 4.3DeepSeek V4 Flash 0731Minimax M3DeepSeek V4 ProDeepSeek V4 Pro 0813Grok Build 0.1DeepSeek V4 Flash| Rank | Model or system | Overall score | Cost / task | Other details | Note |
|---|---|---|---|---|---|
| #1 | Claude Fable 5 Max EffortAnthropic | 83% | $1.439 | — | — |
| #2 | GPT-5.6 Sol Max EffortOpenAI | 81% | $0.515 | — | — |
| #3 | GPT-5.5 Thinking xHigh EffortOpenAI | 80.2% | $0.435 | — | — |
| #4 | Claude 5 Opus Thinking Max EffortAnthropic | 80.1% | $0.699 | — | — |
| #5 | Smaug-AgenticThinking Machines | 79.5% | $0.329 | — | — |
| #6 | Kimi K3Moonshot AI | 79.2% | $0.348 | — | — |
| #7 | Gemini 3.7 Flash HighGoogle | 78.8% | $0.157 | — | — |
| #8 | Qwen 3.8 MaxQwen | 78.5% | $0.275 | — | — |
| #9 | GPT-5.4 Thinking xHigh EffortOpenAI | 78% | $0.387 | — | — |
| #9 | Grok 4.6SpaceXAI | 78% | $0.207 | — | — |
| #9 | Muse Spark 1.2 xHigh EffortMeta | 78% | $0.375 | — | — |
| #12 | GPT-5.6 Terra Max EffortOpenAI | 77.9% | $0.352 | — | — |
| #13 | DeepSeek V4 Pro 0813DeepSeek | 77.4% | $0.044 | — | — |
| #14 | Gemini 3.1 Pro Preview HighGoogle | 77% | $0.286 | — | — |
| #15 | Claude 4.7 Opus Thinking xHigh EffortAnthropic | 76.5% | $0.528 | — | — |
| #16 | Claude 4.8 Opus Thinking Max EffortAnthropic | 76.2% | $0.983 | — | — |
| #17 | Claude Sonnet 5 xHigh EffortAnthropic | 76% | $0.505 | — | — |
| #18 | Grok 4.5SpaceXAI | 75.8% | $0.131 | — | — |
| #19 | Muse Spark 1.1 xHigh EffortMeta | 75.3% | $0.198 | — | — |
| #20 | Gemini 3.5 Flash HighGoogle | 74.6% | $0.249 | — | — |
| #20 | GPT-5.2 HighOpenAI | 74.6% | $0.234 | — | — |
| #22 | Claude 4.6 Opus Thinking High EffortAnthropic | 74.5% | $0.404 | — | — |
| #23 | DeepSeek V4 Flash 0731DeepSeek | 74.2% | $0.060 | — | — |
| #24 | GPT-5.2 CodexOpenAI | 74% | $0.187 | — | — |
| #25 | Gemini 3.6 Flash HighGoogle | 73.6% | $0.235 | — | — |
| #25 | GPT-5.6 Luna Max EffortOpenAI | 73.6% | $0.169 | — | — |
| #27 | GLM-5.2Z.ai | 73.2% | $0.225 | — | — |
| #28 | Qwen 3.7 MaxQwen | 73.1% | $0.182 | — | — |
| #29 | Claude 4.6 Sonnet Thinking Medium EffortAnthropic | 73% | $0.306 | — | — |
| #30 | Claude 4.5 Opus Thinking High EffortAnthropic | 72.6% | $0.610 | — | — |
| #31 | Inkling xHigh EffortInclusionAI | 71.9% | $0.310 | — | — |
| #32 | DeepSeek V4 ProDeepSeek | 71.6% | $0.050 | — | — |
| #33 | Kimi K2.6 ThinkingMoonshot AI | 70.5% | $0.169 | — | — |
| #34 | GPT-5.4 Nano xHighOpenAI | 69.6% | $0.091 | — | — |
| #35 | Qwen 3.6 PlusQwen | 68.9% | $0.227 | — | — |
| #36 | Kimi K2.7 CodeMoonshot AI | 68.4% | $0.100 | — | — |
| #37 | Grok Build 0.1SpaceXAI | 67.8% | $0.024 | — | — |
| #38 | Minimax M3MiniMax | 67.3% | $0.060 | — | — |
| #39 | GPT-5.4 Mini xHighOpenAI | 66.4% | $0.334 | — | — |
| #40 | DeepSeek V4 FlashDeepSeek | 65.5% | $0.016 | — | — |
| #41 | Qwen 3.6 27BQwen | 64% | $0.202 | — | — |
| #42 | Gemini 3.5 Flash-Lite HighGoogle | 63.9% | $0.069 | — | — |
| #43 | Grok 4.3SpaceXAI | 62.3% | $0.061 | — | — |
Full LiveBench 2026-06-25 overall release mirrored from livebench.ai (43 published overall rows with category columns available on the source). LiveBench rotates questions; open the source for category/subtask detail and later releases.
Open LiveBench leaderboardEvidence in this catalog
Where LiveBench overall fits
Closest verified examples we currently carry. A missing score is not a zero.
Model examples
No directly comparable model score is verified here yet.
Harness examples
No directly comparable harness score is verified here yet.